Numerous deaths are triggered by malignant carcinoma annually, which was accounted for 22% of all deaths in 2016 in the United States (Siegel et al., 2019). Immunotherapy, enhancing antitumor immune responses, has provided a new weapon for treating cancer in recent years. Immune checkpoint inhibitors (ICIs), including cytotoxic-T-lymphocyte-associated antigen 4 (CTLA-4), programmed cell death 1 (PD-1) and whose ligand termed programmed cell death ligand 1 (PD-L1), have been approved as therapeutic regimens for various cancers. Of that PD-L1 inhibitors have been confirmed that it could significantly improve the overall survival (OS) and progression-free survival (PFS) of patients with a range of cancers (Brahmer et al., 2012; Rittmeyer et al., 2017). The notable effect of PD-L1 inhibitors indicated it might become the essential component of ICI combination regimens in the future (Kyi and Postow, 2016). Regardless of these benefits, several studies revealed that fatal adverse events associated with PD-L1 inhibitors occasionally emerged in some patients (Barlesi et al., 2018; Choueiri et al., 2018).

The prior studies demonstrated that the common PD-L1 inhibitors-related adverse events were fatigue and diarrhea. The incidence of grade 3 or 4 toxic effects that regarded as related to PD-L1 inhibitors was 9% (Brahmer et al., 2012). Additionally, although rare, previous studies suggested that cardiac toxicity related to PD-L1 inhibitors also occurred in several patients (Varricchi et al., 2018). There were several articles regarding PD-L1 inhibitors reported immune-related adverse events (irAEs), whereas a majority of studies focused on general toxicity instead of lethal toxicity. Regrettably, we have limitedly known about the incidence, the spectrum and susceptible factors of PD-L1 inhibitors-related fatal adverse events until now. Given the increased employment of PD-L1 inhibitors in the clinical setting, the fatal adverse events associated with PD-L1 inhibitors were regarded as a crucial field to be profoundly considered.

Thus, we performed this systematic review and meta-analysis to assess the incidence of fatal adverse events that were associated with the use of PD-L1 inhibitors, to describe them and to statistically depict factors that were associated with these events.

Despite promising progress in the treatment regimen of anti-cancer represented by PD-L1 inhibitors, treatment-related death occasionally emerged. The safety of PD-L1 inhibitors was an essential field to profoundly explore. Therefore, we conducted a meta-analysis involving 26 studies with 6,896 unique participants to elucidate the incidence, the detailed spectrum, and the potential susceptible factors of fatal adverse events associated with PD-L1 inhibitors. The results demonstrated that the overall incidence was 1.24% (95% CI: 0.93–1.65%). In terms of the analysis of spectrum, we observed that ILD was the leading cause of death (n = 12) followed by sepsis (n = 5), neutropenia (n = 4), pneumonia (n = 3), hemoptysis (n = 3), respiratory failure (n = 3), and autoimmune hepatitis (n = 3). Importantly, two rare deaths from PD-L1 inhibitors, relating to myocarditis were documented. As for the analysis of the organs-specific fatal adverse events, the respiratory system (n = 24) was the most frequently reported. According to the results of univariate meta-regression analysis, the potential susceptible factors were as below: the patients treated with PD-L1 inhibitors who involved non-squamous NSCLC, who were included in the trial phase I, and who from the middle-aged group. Additionally, the trial phase of the study was a confirmative predictor based on the results of multivariate meta-regression analysis. Importantly, the aforementioned results, including the incidence, the complete spectrum, the potential susceptible factors, and the relevant predictors of the fatal adverse events associated with PD-L1 inhibitors were firstly reported via searching widespread databases.

There was a meta-analysis displayed the incidence of general adverse events related to PD-1 and PD-L1 inhibitors (Baxi et al., 2018), including fatigue, rash, hepatitis, pneumonitis, and diarrhea. However, our study concentrated on fatal adverse events instead of general adverse events. In addition, our study exhibited the incidence of the fatal adverse events associated with PD-L1 inhibitors was 1.24%, which was remarkably distinct from those reported from the previous meta-analysis incorporated one clinical trial regarding atezolizumab, whose incidence was 0.63% (Abdel-Rahman et al., 2017). The evident distinction might origin from significant different sample sizes (n = 26 vs. n = 1). Also, our results were obviously distinct from the prior study that reported the real-world incidence was 1.97% (613 unique fatal irAEs were identified from 31,059 irAEs case reports) according to the global adverse drug reaction database (Vigilyze-Vigibase) and the pooled incidence was 0.8% based on the pooled meta-analysis (Wang et al., 2018). The distinction might be attributed to different inclusion criteria and sample sizes (n = 26 vs. n = 15). Intriguingly, When the ineligible 44 studies with zero fatal events were added to analyze, the re-evaluated overall incidence slightly altered (1.15%, 95% CI: 0.91–1.45%). However, the inclusion of these studies would introduce substantial bias based on the Begger's test (p < 0.0001).

This study provided the important implications for clinical safety usage of PD-L1 inhibitors. The results revealed that old patients had a higher incidence of fatal adverse events than young patients. The reasonable interpretation of that was the old patients frequently complicated with a series of basic diseases. Consequently, the old patients were more likely to be threatened by fatal adverse events associated with PD-L1 inhibitors, which implied that the clinicians need to monitor old patients more closely when receiving PD-L1 inhibitors.

Moreover, the results also demonstrated that the incidence of patients with non-squamous NSCLC was evidently higher than those with UC. The potential rationale of that might be attributed to the following two factors. One possibility was that a great distinction of PD-L1 expression between NSCLC and UC. Compared with the latter, the former had a remarkable expression of PD-L1 (Patel and Kurzrock, 2015). Besides, partial studies regarding patients with NSCLC or UC revealed that the high levels of PD-L1 expression were related to favorable overall response rate (ORR) and OS (Velcheti et al., 2014; Bellmunt et al., 2015; Brody et al., 2017; Ghate et al., 2019). Further studies exhibited that irAEs were significantly associated with the efficacy of ICIs in patients with NSCLC or melanoma (Hua et al., 2016; Haratani et al., 2018). The above studies seemingly indicated that the PD-L1 expression was positively associated with irAEs. However, we also observed a paradoxical phenomenon that appeared in this study. The incidence in the PD-L1-positive expression group was numerically lower than those in PD-L1-negative expression (0.33 vs. 0.75, p = 0.53). Additionally, a marginally significant difference was discovered in the PD-L1-positive group and mixed ones (0.33 vs. 1.33, p = 0.07). Based on the above contradictory shreds of evidence, further researches were needed to carry out to elucidate the relationship between them. Despite the ambiguity, this study provided a preliminary insight that the therapy of PD-L1 inhibitors in patients with UC was safer than those with non-squamous NSCLC, which indicated that the clinicians should particularly monitor the adverse events occurred in patients with non-squamous NSCLC in case of the emergent of fatal adverse events. Alternatively, the different tumor burden existed in non-squamous NSCLC and UC. The prior study suggested that a high tumor burden was related to serious irAEs in patients with NSCLC (Sakata et al., 2019), which forcefully implied that tumor burden was likely to be an underlying predictor for fatal adverse events associated with PD-L1 inhibitors. There is abundant room for further progress in determining the correlation between them.

The incidence of fatal adverse events associated with PD-L1 inhibitors was higher in phase I than that in phase II according to univariate meta-regression analysis. Intriguingly, after regulating for cancer types and average age, the incidence of phase I had a markedly higher incidence than that in phase II or phase III. The reasonable interpretation was that the design of subsequent trials was optimized based on the experience of early trials to promote the reduction of incidence.

To our knowledge, the predictors of fatal adverse events remained unclear. There were some studies indicated that the occurrence of fatal events might be related to unique genes or specific gastrointestinal flora (Wolchok et al., 2010; Chaput et al., 2017). In this study a total of three potential predisposing factors (non-squamous NSCLC, the middle-aged group, and the trial phase I) based on the results of univariate meta-regression analysis, whereas only a relevant predictor (the trial phase I) was established according to multivariate meta-regression analysis. These easily available clinical characteristics could promote clinicians to quickly predict and identify high-risk populations, and facilitate clinicians to early detect and effectively reduce these thorny issues occurred.

Of note, the amount of respiratory system-related fatal adverse events (n = 24, 32.43%) were predominate across various fatal causes, with ILD being the most prominent (n = 12, 16.22%). ILD, mainly including idiopathic pulmonary fibrosis and pneumonitis in our study, is a group that incorporates more than 200 diseases characterized by diffuse pulmonary parenchymal disease. Beyond that, the increased awareness of the sharp development of circulatory system-related adverse events (e.g. myocarditis), the improved knowledge of extremely rare neurological-related ones (e.g. myasthenia gravis), and the enhanced comprehension of inflammatory infection-related ones (e.g. sepsis and neutropenia) that can be completely reversed by early intervention would develop the safety of using PD-L1 inhibitors in the clinical setting.

PD-L1 inhibitors have shown remarkable antitumor benefit by improving the activity of the immune system via blocking the binding of PD-1 to PD-L1. Unfortunately, however, excessive enhancement of immune system activity could trigger inflammatory side effects, which was termed irAEs. There was some evidence that might explain the underlying mechanism of fatal irAEs. Firstly, homologous antigens existed in some tumors and normal tissues. A report with two fatal myocarditis exhibited that similar T-cell clones were observed in both the tumor and the myocardium (Johnson et al., 2016). Hence the myocardium would be attacked by activated T cells, and developed cardiac-related irAEs. Secondly, IgG-dependent complement fixation and phagocytosis, which would over-activate the immune system, triggering extensive destruction on normal tissue with high CTLA-4 expression (Caturegli et al., 2016) (e.g. pituitary gland). Besides that, the release of substantial cytokines might be involved in the occurrence of irAEs, of which interleukin-17 (IL-17) was the best known. Multiple studies implied that the elevated level of IL-17 was associated with immune-related colitis (Samaan et al., 2018). Additionally, beyond T-cell-mediated immunity, anti-PD-L1 therapy could modulate humoral immunity by increasing levels of preexisting antibodies (Osorio et al., 2017) (e.g. antithyroid antibodies). Despite the precise mechanism, irAEs were attributed to excessive immunity against normal tissue or organs (Postow et al., 2018).

There were several limitations existed in this study. Firstly, this pooled analysis was conducted by using aggregated patient data instead of individual data, which limited the capability to deeply explore the correlation between adverse events and characteristics of patients. Additionally, given the absence of the control group in some studies, the comparison of PD-L1 inhibitors with other drugs could not be performed. Consequently, the pooled analysis merely involved the difference in the incidence of fatal adverse events in diverse PD-L1 inhibitors. Furthermore, given the number of studies regarding PD-L1-positive expression extremely little (n = 2), the role of PD-L1 expression in the incidence of fatal adverse events associated with PD-L1 inhibitors remained controversial.

The overall incidence of fatal adverse events associated with PD-L1 inhibitors was 1.24%, which was markedly higher in patients involved non-squamous NSCLC, trial phase I, and the middle-aged group, compared with the corresponding reference group. The amounts of respiratory system-related fatal adverse events predominated across various fatal causes, of that ILD being the most prominent.

This study firstly provided three potential susceptible factors (non-squamous NSCLC, trial phase I, and the middle-aged group) and ultimately established a relevant predictor (trial phase I) according to multivariate meta-regression analysis, which yielded a capability for clinicians to distinguish high-risk populations from relatively low-risk ones, facilitating to improve the safety of PD-L1 inhibitors which were broadly used in the clinical setting. Importantly, given the maturity of the PD-L1 expression detection methods, we urgently hoped that a variety of studies regarding the correlation between PD-L1 expression and fatal adverse events occurred, prompting clinicians to early detect and effectively handle these thorny issues.

All datasets generated for this study are included in the article/ Supplementary Material .

XW: project design, data analysis, manuscript writing. SW: draft the article, data analysis. YC, ES, TZ, and LL: data collection, data extraction. XC: critically edited the manuscript, supervised the project.

This research received financial support from the National Natural Science Foundation of China (Nos 81502360); the Natural Science Foundation of Fujian Province (Nos 2018J01352, 2016J01576, and 2016J01586); the Science and Technology Innovation Joint Foundation of Fujian Province(Nos 2017Y9125).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.